David Hou (MongoDB): Rethinking Docs as Code in the age of vibe coding

This talk was presented at the AI The Docs 2025 online conference. We are thrilled to share the recording and the summary with you. 

Visit the talk summary page to see all of the presentations from the conference.

presentation

David Hou (Technical Writer at MongoDB)

Summary

How is coding itself fundamentally changing due to AI, particularly with the emergence of "vibe coding" and the rapid adoption of AI-assisted tools by developers? 

In the presentation, David Hou (Technical writer at MongoDB) shares how AI is reshaping software practices and how writers can adapt by treating documentation like code. 

Docs as Code: The Foundation 

David Hou defines docs as code as adopting development practices for documentation: writing in markup, using version control, collaborating via code workflows, and following agile methodologies. This approach treats docs as living, structured artifacts, enabling automation and iterative improvement. The 

Transformation of Coding: Vibe Coding and AI 

He introduces vibe coding, where AI and natural language prompts enable developers to build full applications. AI-assisted coding tools like GitHub Copilot and Cursor are widely adopted, changing what coding looks like, developers now collaborate with AI, relying on structured prompts and rapid iteration. 

Applying AI Lessons to Writing 

If docs are code, writers can adopt similar practices. AI can help them work faster, amplify expertise, and automate repetitive tasks. Writers should treat AI as an assistant, not an author, ensuring human review before publishing. The principles of structured communication, prompt iteration, and workflow integration are crucial. 

AI-Assisted Writing Tools and Best Practices 

• Content Generation: AI can produce code examples, text, API metadata, and markup.
• Review and Compliance: AI can check style, consistency, and correctness.
• Workflow Automation: Agentic AI tools can run scripts to apply changes across documentation.
• Pattern Tasks: AI excels at repeatable tasks, applying consistent changes and reducing manual effort.
• Style Checking & Conversion: LLMs can enforce style guides and convert plain text into structured formats like Markdown or API specs. 

Prompts as Key Artifacts 

David Hou emphasizes that prompts are the most valuable asset. Saving, standardizing, and iterating prompts improves reproducibility, efficiency, and output quality. Providing rich context in prompts enhances AI-generated results. 

Demo Highlights 

• Generating reStructuredText (RST) from natural language outlines.
• Style guide checking for grammar, tense, and voice.
• Agentic testing using MongoDB servers to validate examples and update documentation.
• Automated Pull Requests to integrate changes seamlessly. 

Writing for Humans and LLMs 

David Hou stresses that good documentation for LLMs is largely the same as for humans: clear, correct, structured, and up-to-date. The risk of “garbage in, garbage out” underscores the importance of well-maintained source content. 

MongoDB’s Initiatives and Future Outlook 

MongoDB is transforming docs into knowledge bases for LLMs, integrating vector search, LLM.ext for Markdown, and MCP servers. By providing documentation as structured context, LLMs can generate runnable scripts, reduce hallucinations, and enhance developer productivity. 

Key Takeaways 

Technical writers can: 

• Learn from AI-assisted development.
• Use docs as code as a guiding principle.
• Leverage AI for repetitive and pattern-based tasks.
• Focus on higher-level writing, strategy, and content quality.
• Embrace documentation as a knowledge base for LLMs, ensuring clarity and accuracy for both humans and AI.